AI is introducing a new class of threats that don’t look like traditional attacks and can’t be detected with conventional tools. The AI applications that organizations deploy in the cloud interact with large language models (LLMs) through prompts and responses. This prompt layer has emerged as a new attack surface, where risks like prompt injection and sensitive data leakage can go unnoticed. Prompt injection is now widely recognized as a top risk in AI systems, including in the OWASP Top 10 for
